Gym: Custom Environments Attribute Error

Created on 3 Feb 2018 · 7Comments · Source: openai/gym

I wrote an environment called SimpleEnv that works on the machine I made it on (a couple of weeks ago). I copied the code for this environment onto another machine, installed both it and gym via pip, and I receive this error.

WARN: Environment '<class 'gym_simple.envs.env.SimpleEnv'>' has deprecated methods. Compatibility code invoked.
Traceback (most recent call last):
  File "simple_test.py", line 54, in <module>
    tf.app.run(main)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/platform/app.py", line 124, in run
    _sys.exit(main(argv))
  File "simple_test.py", line 41, in main
    env = gym.make('gym-simple-v0')
  File "/usr/local/lib/python3.5/dist-packages/gym/envs/registration.py", line 163, in make
    return registry.make(id)
  File "/usr/local/lib/python3.5/dist-packages/gym/envs/registration.py", line 121, in make
    patch_deprecated_methods(env)
  File "/usr/local/lib/python3.5/dist-packages/gym/envs/registration.py", line 181, in patch_deprecated_methods
    env.seed  = env._seed
AttributeError: 'SimpleEnv' object has no attribute '_seed'

There was an update to the Github repository between the times of me writing the code and this error occurring. I am unaware if an update was done to the pip gym package in that time, but this seems to be the issue, as the SimpleEnv environment still works on the initial machine (same code)

Source

mvarble

Most helpful comment

@JohnYiQC Hi! I simply changed the methods to no longer have the underscore. So—assuming you are reading the documentation here—you would do the exact same thing, except your gym-foo/gym_foo/envs/foo_env.py file will now be of the form:

import gym
from gym import error, spaces, utils
from gym.utils import seeding

class FooEnv(gym.Env):
  metadata = {'render.modes': ['human']}

  def __init__(self):
    ...
  def step(self, action):
    ...
  def reset(self):
    ...
  def render(self, mode='human', close=False):
    ...

mvarble on 7 Mar 2018

👍4 ❤1

All 7 comments

The new version of gym (0.9.6) assumes that all deprecated envs (that implement the _method() interface instead of the new method() interface), implement a method called _seed(). You can implement a wrapper to your old seed method as _seed() and it should fix the issue.

thevazking on 4 Feb 2018

@thevazking, thank you for the assistance; this worked!

mvarble on 5 Feb 2018

@mvarble , Hi, I'm fairly new to python much less to machine learning, but I encountered exactly the same problem as what you've reported. How exactly did you solve the problem? Can you please post a snippet of your code like how you implemented the wrapper?
Thanks!

JohnYiQC on 6 Mar 2018

👍1

import gym
from gym import error, spaces, utils
from gym.utils import seeding

class FooEnv(gym.Env):
  metadata = {'render.modes': ['human']}

  def __init__(self):
    ...
  def step(self, action):
    ...
  def reset(self):
    ...
  def render(self, mode='human', close=False):
    ...

mvarble on 7 Mar 2018

👍4 ❤1

@mvarble Thanks! That got me going on!

JohnYiQC on 7 Mar 2018

@mvarble and @thevazking : thanks a lot ... :)

avikpram on 7 Mar 2018

I'm sorry, but can someone spell this out for me? I've tried removing the underscore at the front of each method and I still receive the same error when I try to run an env, in this case from https://github.com/xinleipan/gym-gridworld